code-improvement-advisor
Pass
Audited by Gen Agent Trust Hub on Jun 14, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The instructions establish a professional persona for a code review task. There are no patterns suggesting attempts to bypass safety guidelines or override system prompts. All role-play instructions are scoped strictly to the code review use case.
- [DATA_EXFILTRATION]: The skill includes a 'Scope of Analysis' rule that limits the agent to analyzing only snippets or files explicitly named by the user. It also promotes safe practices by listing 'no secrets in source' as a lens for its reviews. There are no network operations or credential harvesting patterns.
- [COMMAND_EXECUTION]: The skill uses file-editing capabilities if provided by the platform. This is an intended feature for applying code improvements. The risk is mitigated by a mandatory two-stage workflow where the agent first proposes changes and then only applies a specific approach after the user explicitly selects it.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process user-provided code files. While this exposes a surface for indirect injection, the risk is managed by the structured response format (tables/detail blocks) and the requirement for a clear user confirmation ('Stage 2') before the agent acts on the interpreted code, preventing the code from silently controlling agent behavior.
Audit Metadata